Wentao Wang1, Dongsheng Gu2,3, Jingwei Wei2,3, Ying Ding1, Li Yang1, Kai Zhu4, Rongkui Luo5, Sheng-Xiang Rao1, Jie Tian6,7,8,9, Mengsu Zeng10. 1. Department of Radiology, Zhongshan Hospital, and Shanghai Medical Imaging Institute, Fudan University, 180 Fenglin Rd, Shanghai, 200032, China. 2. Key Laboratory of Molecular Imaging, Institute of Automation, Chinese Academy of Sciences, 95 Zhongguancun East Road, Beijing, 100190, China. 3. University of Chinese Academy of Sciences, Beijing, 100049, China. 4. Liver Cancer Institute, Zhongshan Hospital, Fudan University, Shanghai, China. 5. Department of Pathology, Zhongshan Hospital, Fudan University, Shanghai, China. 6. Key Laboratory of Molecular Imaging, Institute of Automation, Chinese Academy of Sciences, 95 Zhongguancun East Road, Beijing, 100190, China. tian@ieee.org. 7. University of Chinese Academy of Sciences, Beijing, 100049, China. tian@ieee.org. 8. Beijing Advanced Innovation Center for Big Data-Based Precision Medicine, School of Medicine, Beihang University, Beijing, 100191, China. tian@ieee.org. 9. Engineering Research Center of Molecular and Neuro Imaging of Ministry of Education, School of Life Science and Technology, Xidian University, Xi'an, 710126, Shaanxi, China. tian@ieee.org. 10. Department of Radiology, Zhongshan Hospital, and Shanghai Medical Imaging Institute, Fudan University, 180 Fenglin Rd, Shanghai, 200032, China. zengmengsu@outlook.com.
Abstract
OBJECTIVES: We aimed to develop a radiomics-based model derived from gadoxetic acid-enhanced MR images to preoperatively identify cytokeratin (CK) 19 status of hepatocellular carcinoma (HCC). METHODS: A cohort of 227 patients with single HCC was classified into a training set (n = 159) and a time-independent validated set (n = 68). A total of 647 radiomic features were extracted from multi-sequence MR images. The least absolute shrinkage and selection operator regression and decision tree methods were utilized for feature selection and radiomics signature construction. A multivariable logistic regression model incorporating clinico-radiological features and the fusion radiomics signature was built for prediction of CK19 status by evaluating area under curve (AUC). RESULTS: In the whole cohort, 57 patients were CK19 positive and 170 patients were CK19 negative. By combining 11 and 6 radiomic features extracted in arterial phase and hepatobiliary phase images, respectively, a fusion radiomics signature achieved AUCs of 0.951 and 0.822 in training and validation datasets. The final combined model integrated a-fetoprotein levels, arterial rim enhancement pattern, irregular tumor margin, and the fusion radiomics signature, with a sensitivity of 0.818 and specificity of 0.974 in the training cohort and that of 0.769 and 0.818 in the validated cohort. The nomogram based on the combined model showed satisfactory prediction performance in training (C-index 0.959) and validation (C-index 0.846) dataset. CONCLUSIONS: The combined model based on a fusion radiomics signature derived from arterial and hepatobiliary phase images of gadoxetic acid-enhanced MRI can be a reliable biomarker for CK19 status of HCC. KEY POINTS: • Arterial rim enhancement pattern and irregular tumor margin on hepatobiliary phase on gadoxetic acid-enhanced MRI can be useful for evaluating CK19 status of HCC. • A radiomics-based model performed better than the clinico-radiological model both in training and validation datasets for predicting CK19 status of HCC. • The nomogram based on the fusion radiomics signature can be easily used for CK19 stratification of HCC.
OBJECTIVES: We aimed to develop a radiomics-based model derived from gadoxetic acid-enhanced MR images to preoperatively identify cytokeratin (CK) 19 status of hepatocellular carcinoma (HCC). METHODS: A cohort of 227 patients with single HCC was classified into a training set (n = 159) and a time-independent validated set (n = 68). A total of 647 radiomic features were extracted from multi-sequence MR images. The least absolute shrinkage and selection operator regression and decision tree methods were utilized for feature selection and radiomics signature construction. A multivariable logistic regression model incorporating clinico-radiological features and the fusion radiomics signature was built for prediction of CK19 status by evaluating area under curve (AUC). RESULTS: In the whole cohort, 57 patients were CK19 positive and 170 patients were CK19 negative. By combining 11 and 6 radiomic features extracted in arterial phase and hepatobiliary phase images, respectively, a fusion radiomics signature achieved AUCs of 0.951 and 0.822 in training and validation datasets. The final combined model integrated a-fetoprotein levels, arterial rim enhancement pattern, irregular tumor margin, and the fusion radiomics signature, with a sensitivity of 0.818 and specificity of 0.974 in the training cohort and that of 0.769 and 0.818 in the validated cohort. The nomogram based on the combined model showed satisfactory prediction performance in training (C-index 0.959) and validation (C-index 0.846) dataset. CONCLUSIONS: The combined model based on a fusion radiomics signature derived from arterial and hepatobiliary phase images of gadoxetic acid-enhanced MRI can be a reliable biomarker for CK19 status of HCC. KEY POINTS: • Arterial rim enhancement pattern and irregular tumor margin on hepatobiliary phase on gadoxetic acid-enhanced MRI can be useful for evaluating CK19 status of HCC. • A radiomics-based model performed better than the clinico-radiological model both in training and validation datasets for predicting CK19 status of HCC. • The nomogram based on the fusion radiomics signature can be easily used for CK19 stratification of HCC.